Estimates of leaf area index using Hemispheric photos and MODIS

Transcription

1 Estimates of leaf area index using Hemispheric photos and MODIS William Sea CSIRO Marine and Atmospheric Research Canberra, ACT Ozflux Course, Creswick 5 February 2010 With additional contributions from Philippe Choler, Richard Weinmann, Jason Beringer, Lindsay Hutley, Steve Zegelin, and Ray Leuning

2 Outline for talk Background Remarks on JB Field campaign Results MODIS disasters Summary

3 Background Leaf Area Index is the one-sided green leaf area per unit ground surface area in broadleaf canopies High quality LAI products are needed for water and carbon balance modeling at the regional to continental to global scales Validation of moderate scale remote sensing LAI products are seldom done using ground-based LAI measurements Assessment of MODIS Collection 5 LAI/fPAR products needed for savannas regions of Australia Such validation work presented us with numerous logistical obstacles but also opportunities for initial observations of the vegetation structure and composition Moderate Resolution Imaging Spectroradiometer (MODIS)

4 Renewed interest in DHPs Inexpensive Easy to use (illumination conditions) No reference measurements needed Possible use over low vegetation canopies Direct evaluation of the quality of measurements (images) Possible distinction between green and non-green elements Possible to derive clumping information

5 Advantages of digital hemispheric photos System Illumination conditions Spectral domain Zenith angles Azimuthal coverage Gap size distribution Post processing Computer resources DEMON Direct 430 nm No No Low Sunfleck ceptometer Direct, diffuse PAR Yes Yes Low Accupar Direct, diffuse PAR Yes No Low LAI-2000 Diffuse < 490 nm 5 range No No Low TRAC Direct PAR Yes No Low DHP Direct, diffuse Selectable range range Yes Yes High MVI Diffuse VIS, NIR range range Yes Yes High Ideal Direct, diffuse VIS, NIR range range Yes Yes

6 Clumping of leaves: Unclumped vs. Clumped LAI Effective LAI assumes a Poisson spatial distribution of leaves in the canopy True LAI incorporates a clumping distribution of leaves With a clumped distribution of leaves in the canopy, the LAI is higher than the unclumped LAI, LAI clumped = Ω * LAI unclumped But not all leaves are equal in the canopy!

7 Remarks on LAI Converting from PAI to LAI is not trivial. LAI (as sensed from above) generally has leaves covering stems and trunks in closed canopy forests. For savannas, it all depends on the tree architecture and cover. The clumping coefficient Ω is poorly measured and a large source error in LAI measurements (Weiss et al. 2004). The particular LAI used (in models) depends on the purpose. Represents a potential divide between the measurement community and the others.

8 Reference Maps (Morrisette et al. (2006) To assess the spatial variability of LAI, use data from a high resolution remotely sensed product, e.g. LANDSAT ETM (30 m) Develop relationship between NDVI and LAI Map LAI at 30 meters to compare with ground measurements ~ 900 pixels with 1 MODIS pixel DHP site 30 m

9 CAN-EYE 5.0 Software

10 Perfect to not-so-perfect hemispheric photos

11 Typical Classification using CAN_EYE 5.0 software Unclassified hemispheric photos Two state classification by filling in the sky

12 Essential steps in the process Processing Fieldwork Masking of non vegetation DHPs Subjective classification of green and non-green vegetation Unclumped & clumped estimates of LAI

13 MODIS: from daily surface reflectances to 8-day LAI Surface reflectances (Channels 1, 2, 3-7) & Viewing angles for sensor and zenith 6-biome land cover map Radiative transfer coefficient lookup tables (and backup algorithm) MOD15A2 FPAR, LAI at 1 km 2 Data used from day with max FPAR Backup

14 Currently monthly MODIS LAI

15 Other remotely sensed LAI products CYCLOPS GLOBCARBON ECOCLIMAP AVHRR Several recent papers suggest better performance for CYCLOPS than MODIS (Baret et al. 2008, Garrigues et al. 2008). But, MODIS has a much more friendly user interface than the others. And, MODIS is processed up to date.

16 Savanna Field Campaign 1 September-18 September, 2008 Darwin-Tennant Creek, NT (~900 km) along the Northern Tropical Terrestrial Transect Participants from CSIRO, Monash University, Charles Darwin University, Flinders University, RMIT, and various Europeans Field measurements coordinated with low level aircraft flights measuring CO 2 and H 2 0 fluxes, LIDAR and hyperspectral sensors for vegetation structure, and PLMR for soil moisture (coordination meeting April in Melbourne). We focused our efforts on comparing ground-based measurement of leaf area index with values derived from MODIS Collection 5 LAI/fPAR. This allowed us to actually visit the maximum number of landscapes in the Northern Territory during the campaign.

17 Principal Research Questions How much improved (if any) are LAI estimates using MODIS Collection 5 (MC5) compared with Collection 4 (MC4) in savanna regions of Australia? Is there an LAI offset different from zero at low LAI values? How well does MC5 LAI compare with ground-based estimates derived from hemispheric photos? Does clumping matter? What is the pattern of LAI along the NATT?

18 Field sites and rainfall gradient

19 MODIS pixels & sampling MODIS footprint Photographer: Steve Zegelin Field sampling Photos taken ~ every 20 meters along transects

20 Result 1: Comparison of MODIS Collection 4 and 5 LAI Mean +/- 1 SE Sea et al. (2009) Remote Sensing of Environment in revision

21 Result 2: Stem LAI ~ 20% of total LAI Sea et al. (2009) Remote Sensing of Environment in revision

22 Result 3: Comparison of MODIS to hemispheric photos For clumped LAI, SMA regression, r 2 = 0.89, error dominated by unsystematic component Error bars are 95 % confidence intervals Sea et al. (2009) Remote Sensing of Environment in revision

23 Result 4: Leaf area index where there should be none! Table 1. LAI offset sampling sites in the Northern Territory. Site Latitude Longitude Description Ground LAI MODIS LAI Bare Senescent Senescent Bare Senescent Mean = 0.28 Std = 0.13 Sea et al. (2009) Remote Sensing of Environment in revision

24 Result 5: LAI along the rainfall gradient Day 89 Day 257 Darwin 1700 mm MAP 12.5 o S MODIS Collection 5 LAI (m 2 /m 2 ) Tennant Creek 19.0 o S 550 mm Latitude South Data via DAAC ONRL MODIS website

25 Potential Errors in MODIS LAI Errors in the radiances Incorrect biome classification Inherent problems of treating classes as homogeneous, e.g. deciduous African savannas & evergreen Australian savannas Structural problems in the model, e.g. reliance on backup model

26 MODIS disasters: Tumbarumba & Otway: 8 7 Tumbarumba Land Cover Evergreen Broadleaf Forest (green) MODIS Collection 5 LAI (m 2 /m 2 ) Otway 7 MODIS Collection 5 LAI (m 2 /m 2 )

27 Summary The choice of equipment for LAI measurement depends on site and research requirements, and budget. Beware of the caveats, especially the difficulties in measuring clumping index. Our results show that MODIS Collection 5 does a reasonably good job at estimating LAI and compares well with DHPs Our results suggest that DHPs should incorporate clumping for comparison to MODIS LAI OR that MODIS is able to capture clumping of vegetation. Some disasters remain for MODIS LAI, with estimates not passing the Leuning Laugh Test (LLT).